Impact of different variables on the outcome of patients with clinically confined prostate carcinoma: prediction of pathologic stage and biochemical failure using an artificial neural network.

BACKGROUND: The advent of advanced computing techniques has provided the opportunity to analyze clinical data using artificial intelligence techniques. This study was designed to determine whether a neural network could be developed using preoperative prognostic indicators to predict the pathologic stage and time of biochemical failure for patients who undergo radical prostatectomy. METHODS: The preoperative information included TNM stage, prostate size, prostate specific antigen (PSA) level, biopsy results (Gleason score and percentage of positive biopsy), as well as patient age. All 309 patients underwent radical prostatectomy at the University of Colorado Health Sciences Center. The data from all patients were used to train a multilayer perceptron artificial neural network. The failure rate was defined as a rise in the PSA level > 0.2 ng/mL. The biochemical failure rate in the data base used was 14.2%. Univariate and multivariate analyses were performed to validate the results. RESULTS: The neural network statistics for the validation set showed a sensitivity and specificity of 79% and 81%, respectively, for the prediction of pathologic stage with an overall accuracy of 80% compared with an overall accuracy of 67% using the multivariate regression analysis. The sensitivity and specificity for the prediction of failure were 67% and 85%, respectively, demonstrating a high confidence in predicting failure. The overall accuracy rates for the artificial neural network and the multivariate analysis were similar. CONCLUSIONS: Neural networks can offer a convenient vehicle for clinicians to assess the preoperative risk of disease progression for patients who are about to undergo radical prostatectomy. Continued investigation of this approach with larger data sets seems warranted.

Evaluation of artificial neural networks for the prediction of pathologic stage in prostate carcinoma.

BACKGROUND: Currently, the standard for predicting pathologic stage from information available at the time of prostate biopsy is the "Partin nomograms" that were derived using logistic regression analysis. The authors retrospectively reviewed a large series of men with clinically localized prostate carcinoma who underwent staging pelvic lymphadenectomy and radical retropubic prostatectomy. They then utilized pathologic and clinical data at the time of prostate biopsy to develop and test an artificial neural network (ANN) to predict the final pathologic stage for this group of men. They then compared the results of ANN with the previous nomograms. METHODS: Five thousand seven hundred forty-four men were treated at the authors' institution from 1985 to 1998. An ANN was developed using two randomly selected training and validation sets for predicting pathologic stage. Input variables included age, preoperative serum prostate specific antigen level, clinical TNM (tumor, lymph node, and metastasis) classification, and Gleason score from the biopsy specimen. Outcomes included organ confinement and lymph node involvement status. RESULTS: The ANN was slightly superior to the nomograms in predicting pathologic stage, such as organ confinement and lymph node involvement status. CONCLUSIONS: In predicting organ confinement and lymph node involvement status, ANN was more accurate and had a larger area under ROC than the nomograms based on the logistic regression method. Artificial neural network models can be developed and used to better predict final pathologic stage when preoperative pathologic and clinical features are known.

The problem of cutoff levels in a screened population: appropriateness of informing screenees about their risk of having prostate carcinoma.

BACKGROUND: Transrectal prostate biopsy decisions often have been based on absolute cutoff values for total and free prostate-specific antigen (PSA). The authors decided that it would be more appropriate to develop risk profiles for the individual patient to allow him to decide whether to undergo a prostate biopsy. METHODS: To develop risk profiles, the authors first used multivariate logistic regression analysis to analyze 2054 males who were part of the Tyrol (Austria) PSA Screening Project. Second, artificial neural network (ANN) analyses were performed using data from 3474 males who also were part of the Tyrol PSA Screening Project and who had undergone prostate biopsy. These analyses were compared with standard cutoff levels of specificity for the detection of prostate carcinoma. RESULTS: To the authors' knowledge, this was the first time that multivariate logistic regression analysis was used to decide whether to perform prostate biopsies based on risk profiles rather than on single cutoff levels. For the detection of prostate carcinoma, at sensitivity levels of 90--95%, the ANN was 150--200% more specific than the standard cutoff points. For screened volunteers with total PSA levels below 4 ng/mL, ANN showed a lower cancer predictive ability in comparison with volunteers with total PSA levels above 4 ng/mL. However, the ANN was approximately 150--200% more specific than the standard cutoff levels in both groups. CONCLUSIONS: At high sensitivity levels, ANN increased the specificity for prostate carcinoma detection in a PSA-based screened population. The improvement in specificity between standard cutoff levels and ANN ranged between 150--200% and was not affected by the presence of benign prostatic hyperplasia or prostatitis.

Neural network and regression predictions of 5-year survival after colon carcinoma treatment.

BACKGROUND: The Commission on Cancer data from the National Cancer Data Base (NCDB) for patients with colon carcinoma was used to develop several artificial neural network and regression-based models. These models were designed to predict the likelihood of 5-year survival after primary treatment for colon carcinoma. METHODS: Two modeling methods were used in the study. Artificial neural networks were used to select the more important variables from the NCDB database and model 5-year survival. A standard parametric logistic regression also was used to model survival and the two methods compared on a prospective set of patients not used in model development. RESULTS: The neural network yielded a receiver operating characteristic (ROC) area of 87.6%. At a sensitivity to mortality of 95% the specificity was 41%. The logistic regression yielded a ROC area of 82% and at a sensitivity to mortality of 95% gave a specificity of 27%. CONCLUSIONS: The neural network found a strong pattern in the database predictive of 5-year survival status. The logistic regression produced somewhat less accurate, but good, results.

Introduction to artificial neural networks for physicians: taking the lid off the black box.

BACKGROUND: Over the past 5 years, a steady stream of publications has discussed the use of artificial neural networks (ANNs) for urologic and other medical applications. The pace of this research has increased recently, and deployed products based on this technology are now appearing. Before these tools can be widely accepted by clinicians and researchers, a deeper level of understanding of ANNs is necessary. This article attempts to lay some of the groundwork needed to facilitate this familiarity. METHODS: A short discussion of neural network history is included for background. This is followed by an in-depth discussion of how and why ANNs work. This discussion includes the relationship between ANNs and statistical regression. An investigation of issues associated with neural networks follows, applicable to both general and urologic-specific applications. RESULTS: Neural networks are computer models that have been studied extensively for over 50 years, with prostate cancer applications since 1994. From a biological viewpoint, ANNs are artificial analogues of data structures that exist in nervous systems. From a numeric viewpoint, ANNs are matrices of numbers whose values comprise knowledge that is distilled from historic databases. Many types of neural networks are analogous to well-known statistical methods. CONCLUSIONS: ANNs are complex numeric constructs, but no more complex than similar statistical methods. However, several issues associated with neural network derivation demand that developers apply rigorous engineering practices in their studies.

Neural network analysis of the prediction of cancer recurrence following debulking laparotomy and chemotherapy in stages III and IV ovarian cancer.

An artificial neural network (ANN) has been developed to predict the presence or absence of cancer following debulking laparotomy and chemotherapy in patients with stages III and IV ovarian cancer. The presence or absence of a residual gross tumor or microscopic disease was determined by a second-look laparotomy. The ANN was trained and tested using detailed operative findings and related surgical procedures associated with the debulking surgery. The ANN predictive results were compared with linear and logistic regression. The ANN significantly outperformed both logistic and linear regression analyses, but additional cases are needed to validate the network.

A neural network predicts progression for men with gleason score 3+4 versus 4+3 tumors after radical prostatectomy.

OBJECTIVES: To determine the significance of Gleason scores 3+4 (GS3+4) versus 4+3 (GS4+3) with respect to biochemical recurrence in a retrospective review of a series of men with clinically localized prostate cancer who underwent radical retropubic prostatectomy (RRP) and to develop and test an artificial neural network (ANN) to predict the biochemical recurrence after surgery for this group of men using the pathologic and clinical data. METHODS: From 1982 to 1998, 600 men had pathologic Gleason score 7 disease without lymph node or seminal vesicle involvement. We analyzed the freedom from biochemical (prostate-specific antigen) progression after RRP on 564 of these men on the basis of their GS3+4 versus GS4+3 (Gleason 7) status. The Cox proportional hazards model was used to determine the importance of Gleason 7 status as an independent predictor of progression. In addition, an ANN was developed using randomly selected training and validation sets for predicting biochemical recurrence at 3 or 5 years. Different input variable subsets, with or without Gleason 7 status, were compared for the ability of the ANN to maximize the prediction of progression. Standard logistic regression was used concurrently on the same random patient population sets to calculate progression risk. RESULTS: A significant recurrence-free survival advantage was found in men who underwent RRP for GS3+4 compared with those with GS4+3 disease (P <0.0001). The ANN, logistic regression, and proportion hazard models demonstrated the importance of Gleason 7 status in predicting patient outcome. The ANN was better than logistic regression in predicting patient outcome, in terms of prostate-specific antigen progression, at 3 and 5 years. CONCLUSIONS: A simple modification of the Gleason scoring system for men with Gleason 7 disease revealed a difference in the patient outcome after RRP. ANN models can be developed and used to better predict patient outcome when pathologic and clinical features are known.

Evaluation of prostate cancer patients receiving multiple staging tests, including ProstaScint scintiscans.

BACKGROUND: Multiple serum tests were performed on archival samples from patients who participated in trials to assess the ProstaScint scan staging ability. Traditional statistical analysis as well as artificial neural network (ANN) analysis were employed to evaluate individual patients and the group as a whole. The results were evaluated so that each factor was tested for prognostic value. METHODS: Data obtained from serum tests, bone scans, and ProstaScint scans were evaluated by traditional statistical methods and ANN to determine the individual value in clinical staging of prostate cancer. RESULTS: Two hundred seventy-five patients (180 postprostatectomy, 95 intact prostate) with prostate cancer (14 with distant metastases) were available for analysis. Data available included: clinical state (remission or progression), most recent clinical TNM stage, bone scan, and ProstaScint scan. Serum was tested for prostate-specific membrane antigen(PSMA), prostate-specific antigen(PSA), free PSA (fPSA), and complexed PSA (cPSA). Additional calculations included percent free PSA, and percent complexed PSA. Spearman individual statistical assessment for traditional group evaluation revealed no significant factors for T-stage. The free PSA and complex PSA had a significant association with node (N)-status. The distant metastases (M) stage correlated well with the bone scan and clinical stage. ANN analysis revealed no significant T-stage factors. N-stage factors showed a 95% sensitivity and 49% specificity. These factors included the presence or absence of a prostate, PSA serum levels, bone scan, and ProstaScint scans as major associated indicators. ANN analysis of the important variables for M-stage included ProstaScint scan score, and PSA levels (total, percent complexed, percent free, and fPSA). These factors were associated with a 95% sensitivity and 15% specificity level. CONCLUSIONS: Two hundred seventy-five patients receiving treatment for prostate cancer were evaluated by ANN and traditional statistical analysis for factors related to stage of disease. ANN revealed that PSA levels, determined by a variety of ways, ProstaScint scan, and bone scan, were significant variables that had prognostic value in determining the likelihood of nodal disease, or distant disease in prostate cancer patients.

Neural network prediction of prostate tissue composition based on magnetic resonance imaging analysis. A pilot study.

OBJECTIVE: To develop a neural network model that estimates prostate histology using magnetic resonance imaging (MRI). STUDY DESIGN: Fifty-three men with lower urinary tract symptoms (average age = 63.8 +/- 8.9 years) underwent a prostate MRI (T2) and sextant biopsy of the prostate. Masson Trichome and immunohistochemical prostate-specific antigen staining of the biopsy material were used to calculate the amount of stroma and epithelium in the inner gland (central plus transition zone). MRIs were normalized to the mean intensity of the obturator internus muscle for comparative analyses. Gray scale and texture features were extracted from the inner gland in the midsection transverse MRI slice. Clinical and image variables were used in two neural networks predicting a high amount of stroma and a high amount of epithelium, respectively. RESULTS: The positive and negative predictive values of the stroma and epithelium neural networks were 95%, 69% and 65%, 92%, respectively. CONCLUSION: These data suggest that the combined use of these neural networks may predict patient response to medical therapy targeting prostatic stroma or epithelium.

Artificial neural networks in urology: Update 2000.

Artificial neural networks (ANNs) are widely available and have been demonstrated to be superior to standard empirical methods of detecting, staging and monitoring prostate cancer. These algorithms have been statistically validated in diverse, well-characterized patient groups and are now being evaluated for clinical use worldwide. New variables based on demographic data, tissue and serum markers show promise for improving our ability to predict disease extent and outcome and may be integrated in future ANN models. This review focuses on recently developed neural networks for detecting, staging and monitoring prostate cancer.

Ten-year disease free survival after transperineal sonography-guided iodine-125 brachytherapy with or without 45-gray external beam irradiation in the treatment of patients with clinically localized, low to high Gleason grade prostate carcinoma.

BACKGROUND: The authors report observed 10-year brachytherapy results in the treatment of 152 consecutive patients with clinically organ-confined prostate carcinoma. METHODS: One hundred and fifty-two consecutive patients with T1-T3, low to high Gleason grade, prostate carcinoma were treated between January 1987 and June 1988 at Northwest Hospital in Seattle, Washington. Their median age was 70 years (range, 53-92 years). Of these 152 patients, 98 (64%) received an iodine-125 implant alone (Group 1), and the remaining 54 patients (36%), who were judged to have a higher risk of extraprostatic extension, also were treated with 45 gray (Gy) of external beam irradiation to the pelvis (Group 2). No patient underwent lymph node sampling, and none received androgen ablation therapy. Multivariate regression and the Mann-Whitney rank sum test were used for statistical analysis. Preoperative patient data with associated success or failure outcomes at 10 years after treatment were used for training and validating a back-propagation neural network prediction program. RESULTS: The average preoperative prostate specific antigen (PSA) value, clinical stage, and Gleason grade were 11.0 ng/mL, T2, and 5, respectively. The median posttreatment follow-up was 119 months (range, 3-134 months). Overall survival 10 years after treatment was 65%. At last follow-up only 3 of the 152 patients (2%) had died of prostate carcinoma. Ninety-seven patients (64%) remained clinically and biochemically free of disease at 10 years of follow-up and had an average PSA value of 0.18 ng/mL (range, 0.01-0.5 ng/mL). In these patients a period of 42 months was required to reach the average PSA (0.5 ng/mL). The median to last PSA follow-up was 95 months (range, 3-134 months). Postoperative needle biopsies were negative in 56% of patients, positive in 15% of patients, and not available in 29% of patients. Only 6% of patients developed bone metastasis. At 10 years there was no statistically significant difference in treatment outcome between patients who received iodine-125 alone, and those who received iodine-125 with 45-Gy external beam irradiation (P = 0.08). Nevertheless, in these two groups preoperative PSA, stage, and Gleason grade were significantly different (P < 0.01). In the artificial neural network analysis, pretreatment serum PSA was the most accurate predictor of disease-free survival. CONCLUSIONS: Percutaneous prostate brachytherapy is a valid and efficient option for treating patients with clinically organ-confined, low to high Gleason grade, prostate carcinoma. Observed 10-year follow-up documents serum PSA levels superior to those reported in several published external beam irradiation series, and comparable to those published in a number of published radical prostatectomy series.

Artificial neural networks: current status in cardiovascular medicine.

Artificial neural networks are a form of artificial computer intelligence that have been the subject of renewed research interest in the last 10 years. Although they have been used extensively for problems in engineering, they have only recently been applied to medical problems, particularly in the fields of radiology, urology, laboratory medicine and cardiology. An artificial neural network is a distributed network of computing elements that is modeled after a biologic neural system and may be implemented as a computer software program. It is capable of identifying relations in input data that are not easily apparent with current common analytic techniques. The functioning artificial neural network's knowledge is built on learning and experience from previous input data. On the basis of this prior knowledge, the artificial neural network can predict relations found in newly presented data sets. In cardiology, artificial neural networks have been successfully applied to problems in the diagnosis and treatment of coronary artery disease and myocardial infarction, in electrocardiographic interpretation and detection of arrhythmias and in image analysis in cardiac radiography and sonography. This report focuses on the current status of artificial neural network technology in cardiovascular medical research.

Neural network analysis of quantitative histological factors to predict pathological stage in clinical stage I nonseminomatous testicular cancer.

A great deal of controversy exists in staging clinical stage I (CSI) nonseminomatous testicular germ cell tumors (NSGCT) because of the difficulty of distinguishing true stage I patients from those with occult retroperitoneal or distant metastases. The goal of this study was to quantitate primary tumor histologic factors and to apply these in a neural network computer analysis to determine if more accurate staging could be achieved. All available primary tumor histological slides from 93 CSI NSGCT patients were analyzed for vascular invasion (VI), lymphatic invasion (LI), tunical invasion (TI) and quantitative determination of percentage of the primary tumor composed of embryonal carcinoma (%EMB), yolk sac carcinoma (%YS), teratoma (%TER) and seminoma (%SEM). These patients had undergone retroperitoneal lymphadenectomy or follow-up such that final stage included 55 pathologic stage I and 38 stage II or higher lesions. Two investigators were provided identical datasets for neural network analysis; one experienced researcher used custom Kohonen and back propagation programs and one less experienced researcher used a commercially available program. For each experiment, a subset of data was used for training, and subsets were blindly used to test the accuracy of the networks. In the custom back propagation network, 86 of 93 patients were correctly staged for an overall accuracy of 92% (sensitivity 88%, specificity 96%). Using Neural Ware commercial software 74 of 93 (79.6%) were accurately staged when all 7 input variables were used; however, accuracy improved from 84.9 to 87.1% when 2, 4 and 5 of the variables were used. Quantitative histologic assessment of the primary tumor and neural network processing of data may provide clinically useful information in the CSI NSGCT population; however, the expertise of the network researcher appears to be important, and commercial software in general use may not be superior to standard regression analysis. Prospective testing of expert methodology should be instituted to confirm its utility.

Artificial neural networks in the diagnosis and prognosis of prostate cancer: a pilot study.

There is controversy about how prostate cancer screening tests should best be used because of the false-negative and false-positive results. There also is controversy about prostate cancer treatment because of errors in tumor staging, uncertainty about treatment efficacy and the variable natural history of the disease. We sought to determine in a pilot study whether artificial neural networks would be helpful to predict biopsy results in men with abnormal screening test(s) and to predict treatment outcome after radical prostatectomy. To predict biopsy results, we extracted data from a prostate specific antigen (PSA) based screening study data base in 1,787 men with a serum PSA concentration of more than 4.0 ng./ml. (approximately 40% of the men also had suspicious findings on digital rectal examination). To predict cancer recurrence after radical prostatectomy, we extracted data from a random sample of 240 patients selected from a data base of men who had undergone radical prostatectomy. The neural network predicted the biopsy result with 87% overall accuracy, and its output threshold could be adjusted to achieve the desired tradeoff between sensitivity and specificity. It also predicted tumor recurrence with 90% overall accuracy. We conclude that trained neural networks may be useful in decision making for prostate cancer patients.